FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_mmap.c
1 /*-
2 * Copyright (c) 1988 University of Utah.
3 * Copyright (c) 1991, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * the Systems Programming Group of the University of Utah Computer
8 * Science Department.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * from: Utah $Hdr: vm_mmap.c 1.6 91/10/21$
35 *
36 * @(#)vm_mmap.c 8.4 (Berkeley) 1/12/94
37 */
38
39 /*
40 * Mapped file (mmap) interface to VM
41 */
42
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD: releng/9.0/sys/vm/vm_mmap.c 225617 2011-09-16 13:58:51Z kmacy $");
45
46 #include "opt_compat.h"
47 #include "opt_hwpmc_hooks.h"
48
49 #include <sys/param.h>
50 #include <sys/systm.h>
51 #include <sys/capability.h>
52 #include <sys/kernel.h>
53 #include <sys/lock.h>
54 #include <sys/mutex.h>
55 #include <sys/sysproto.h>
56 #include <sys/filedesc.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/racct.h>
60 #include <sys/resource.h>
61 #include <sys/resourcevar.h>
62 #include <sys/vnode.h>
63 #include <sys/fcntl.h>
64 #include <sys/file.h>
65 #include <sys/mman.h>
66 #include <sys/mount.h>
67 #include <sys/conf.h>
68 #include <sys/stat.h>
69 #include <sys/sysent.h>
70 #include <sys/vmmeter.h>
71
72 #include <security/mac/mac_framework.h>
73
74 #include <vm/vm.h>
75 #include <vm/vm_param.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vm_pageout.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_page.h>
84
85 #ifdef HWPMC_HOOKS
86 #include <sys/pmckern.h>
87 #endif
88
89 #ifndef _SYS_SYSPROTO_H_
90 struct sbrk_args {
91 int incr;
92 };
93 #endif
94
95 static int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
96 int *, struct vnode *, vm_ooffset_t *, vm_object_t *);
97 static int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
98 int *, struct cdev *, vm_ooffset_t *, vm_object_t *);
99 static int vm_mmap_shm(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *,
100 int *, struct shmfd *, vm_ooffset_t, vm_object_t *);
101
102 /*
103 * MPSAFE
104 */
105 /* ARGSUSED */
106 int
107 sys_sbrk(td, uap)
108 struct thread *td;
109 struct sbrk_args *uap;
110 {
111 /* Not yet implemented */
112 return (EOPNOTSUPP);
113 }
114
115 #ifndef _SYS_SYSPROTO_H_
116 struct sstk_args {
117 int incr;
118 };
119 #endif
120
121 /*
122 * MPSAFE
123 */
124 /* ARGSUSED */
125 int
126 sys_sstk(td, uap)
127 struct thread *td;
128 struct sstk_args *uap;
129 {
130 /* Not yet implemented */
131 return (EOPNOTSUPP);
132 }
133
134 #if defined(COMPAT_43)
135 #ifndef _SYS_SYSPROTO_H_
136 struct getpagesize_args {
137 int dummy;
138 };
139 #endif
140
141 /* ARGSUSED */
142 int
143 ogetpagesize(td, uap)
144 struct thread *td;
145 struct getpagesize_args *uap;
146 {
147 /* MP SAFE */
148 td->td_retval[0] = PAGE_SIZE;
149 return (0);
150 }
151 #endif /* COMPAT_43 */
152
153
154 /*
155 * Memory Map (mmap) system call. Note that the file offset
156 * and address are allowed to be NOT page aligned, though if
157 * the MAP_FIXED flag it set, both must have the same remainder
158 * modulo the PAGE_SIZE (POSIX 1003.1b). If the address is not
159 * page-aligned, the actual mapping starts at trunc_page(addr)
160 * and the return value is adjusted up by the page offset.
161 *
162 * Generally speaking, only character devices which are themselves
163 * memory-based, such as a video framebuffer, can be mmap'd. Otherwise
164 * there would be no cache coherency between a descriptor and a VM mapping
165 * both to the same character device.
166 */
167 #ifndef _SYS_SYSPROTO_H_
168 struct mmap_args {
169 void *addr;
170 size_t len;
171 int prot;
172 int flags;
173 int fd;
174 long pad;
175 off_t pos;
176 };
177 #endif
178
179 /*
180 * MPSAFE
181 */
182 int
183 sys_mmap(td, uap)
184 struct thread *td;
185 struct mmap_args *uap;
186 {
187 #ifdef HWPMC_HOOKS
188 struct pmckern_map_in pkm;
189 #endif
190 struct file *fp;
191 struct vnode *vp;
192 vm_offset_t addr;
193 vm_size_t size, pageoff;
194 vm_prot_t cap_maxprot, prot, maxprot;
195 void *handle;
196 objtype_t handle_type;
197 int flags, error;
198 off_t pos;
199 struct vmspace *vms = td->td_proc->p_vmspace;
200 cap_rights_t rights;
201
202 addr = (vm_offset_t) uap->addr;
203 size = uap->len;
204 prot = uap->prot & VM_PROT_ALL;
205 flags = uap->flags;
206 pos = uap->pos;
207
208 fp = NULL;
209
210 /* Make sure mapping fits into numeric range, etc. */
211 if ((uap->len == 0 && !SV_CURPROC_FLAG(SV_AOUT) &&
212 curproc->p_osrel >= P_OSREL_MAP_ANON) ||
213 ((flags & MAP_ANON) && (uap->fd != -1 || pos != 0)))
214 return (EINVAL);
215
216 if (flags & MAP_STACK) {
217 if ((uap->fd != -1) ||
218 ((prot & (PROT_READ | PROT_WRITE)) != (PROT_READ | PROT_WRITE)))
219 return (EINVAL);
220 flags |= MAP_ANON;
221 pos = 0;
222 }
223
224 /*
225 * Align the file position to a page boundary,
226 * and save its page offset component.
227 */
228 pageoff = (pos & PAGE_MASK);
229 pos -= pageoff;
230
231 /* Adjust size for rounding (on both ends). */
232 size += pageoff; /* low end... */
233 size = (vm_size_t) round_page(size); /* hi end */
234
235 /*
236 * Check for illegal addresses. Watch out for address wrap... Note
237 * that VM_*_ADDRESS are not constants due to casts (argh).
238 */
239 if (flags & MAP_FIXED) {
240 /*
241 * The specified address must have the same remainder
242 * as the file offset taken modulo PAGE_SIZE, so it
243 * should be aligned after adjustment by pageoff.
244 */
245 addr -= pageoff;
246 if (addr & PAGE_MASK)
247 return (EINVAL);
248
249 /* Address range must be all in user VM space. */
250 if (addr < vm_map_min(&vms->vm_map) ||
251 addr + size > vm_map_max(&vms->vm_map))
252 return (EINVAL);
253 if (addr + size < addr)
254 return (EINVAL);
255 } else {
256 /*
257 * XXX for non-fixed mappings where no hint is provided or
258 * the hint would fall in the potential heap space,
259 * place it after the end of the largest possible heap.
260 *
261 * There should really be a pmap call to determine a reasonable
262 * location.
263 */
264 PROC_LOCK(td->td_proc);
265 if (addr == 0 ||
266 (addr >= round_page((vm_offset_t)vms->vm_taddr) &&
267 addr < round_page((vm_offset_t)vms->vm_daddr +
268 lim_max(td->td_proc, RLIMIT_DATA))))
269 addr = round_page((vm_offset_t)vms->vm_daddr +
270 lim_max(td->td_proc, RLIMIT_DATA));
271 PROC_UNLOCK(td->td_proc);
272 }
273 if (flags & MAP_ANON) {
274 /*
275 * Mapping blank space is trivial.
276 */
277 handle = NULL;
278 handle_type = OBJT_DEFAULT;
279 maxprot = VM_PROT_ALL;
280 cap_maxprot = VM_PROT_ALL;
281 } else {
282 /*
283 * Mapping file, get fp for validation and don't let the
284 * descriptor disappear on us if we block. Check capability
285 * rights, but also return the maximum rights to be combined
286 * with maxprot later.
287 */
288 rights = CAP_MMAP;
289 if (prot & PROT_READ)
290 rights |= CAP_READ;
291 if ((flags & MAP_SHARED) != 0) {
292 if (prot & PROT_WRITE)
293 rights |= CAP_WRITE;
294 }
295 if (prot & PROT_EXEC)
296 rights |= CAP_MAPEXEC;
297 if ((error = fget_mmap(td, uap->fd, rights, &cap_maxprot,
298 &fp)) != 0)
299 goto done;
300 if (fp->f_type == DTYPE_SHM) {
301 handle = fp->f_data;
302 handle_type = OBJT_SWAP;
303 maxprot = VM_PROT_NONE;
304
305 /* FREAD should always be set. */
306 if (fp->f_flag & FREAD)
307 maxprot |= VM_PROT_EXECUTE | VM_PROT_READ;
308 if (fp->f_flag & FWRITE)
309 maxprot |= VM_PROT_WRITE;
310 goto map;
311 }
312 if (fp->f_type != DTYPE_VNODE) {
313 error = ENODEV;
314 goto done;
315 }
316 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
317 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
318 /*
319 * POSIX shared-memory objects are defined to have
320 * kernel persistence, and are not defined to support
321 * read(2)/write(2) -- or even open(2). Thus, we can
322 * use MAP_ASYNC to trade on-disk coherence for speed.
323 * The shm_open(3) library routine turns on the FPOSIXSHM
324 * flag to request this behavior.
325 */
326 if (fp->f_flag & FPOSIXSHM)
327 flags |= MAP_NOSYNC;
328 #endif
329 vp = fp->f_vnode;
330 /*
331 * Ensure that file and memory protections are
332 * compatible. Note that we only worry about
333 * writability if mapping is shared; in this case,
334 * current and max prot are dictated by the open file.
335 * XXX use the vnode instead? Problem is: what
336 * credentials do we use for determination? What if
337 * proc does a setuid?
338 */
339 if (vp->v_mount != NULL && vp->v_mount->mnt_flag & MNT_NOEXEC)
340 maxprot = VM_PROT_NONE;
341 else
342 maxprot = VM_PROT_EXECUTE;
343 if (fp->f_flag & FREAD) {
344 maxprot |= VM_PROT_READ;
345 } else if (prot & PROT_READ) {
346 error = EACCES;
347 goto done;
348 }
349 /*
350 * If we are sharing potential changes (either via
351 * MAP_SHARED or via the implicit sharing of character
352 * device mappings), and we are trying to get write
353 * permission although we opened it without asking
354 * for it, bail out.
355 */
356 if ((flags & MAP_SHARED) != 0) {
357 if ((fp->f_flag & FWRITE) != 0) {
358 maxprot |= VM_PROT_WRITE;
359 } else if ((prot & PROT_WRITE) != 0) {
360 error = EACCES;
361 goto done;
362 }
363 } else if (vp->v_type != VCHR || (fp->f_flag & FWRITE) != 0) {
364 maxprot |= VM_PROT_WRITE;
365 cap_maxprot |= VM_PROT_WRITE;
366 }
367 handle = (void *)vp;
368 handle_type = OBJT_VNODE;
369 }
370 map:
371 td->td_fpop = fp;
372 maxprot &= cap_maxprot;
373 error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
374 flags, handle_type, handle, pos);
375 td->td_fpop = NULL;
376 #ifdef HWPMC_HOOKS
377 /* inform hwpmc(4) if an executable is being mapped */
378 if (error == 0 && handle_type == OBJT_VNODE &&
379 (prot & PROT_EXEC)) {
380 pkm.pm_file = handle;
381 pkm.pm_address = (uintptr_t) addr;
382 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
383 }
384 #endif
385 if (error == 0)
386 td->td_retval[0] = (register_t) (addr + pageoff);
387 done:
388 if (fp)
389 fdrop(fp, td);
390
391 return (error);
392 }
393
394 int
395 freebsd6_mmap(struct thread *td, struct freebsd6_mmap_args *uap)
396 {
397 struct mmap_args oargs;
398
399 oargs.addr = uap->addr;
400 oargs.len = uap->len;
401 oargs.prot = uap->prot;
402 oargs.flags = uap->flags;
403 oargs.fd = uap->fd;
404 oargs.pos = uap->pos;
405 return (sys_mmap(td, &oargs));
406 }
407
408 #ifdef COMPAT_43
409 #ifndef _SYS_SYSPROTO_H_
410 struct ommap_args {
411 caddr_t addr;
412 int len;
413 int prot;
414 int flags;
415 int fd;
416 long pos;
417 };
418 #endif
419 int
420 ommap(td, uap)
421 struct thread *td;
422 struct ommap_args *uap;
423 {
424 struct mmap_args nargs;
425 static const char cvtbsdprot[8] = {
426 0,
427 PROT_EXEC,
428 PROT_WRITE,
429 PROT_EXEC | PROT_WRITE,
430 PROT_READ,
431 PROT_EXEC | PROT_READ,
432 PROT_WRITE | PROT_READ,
433 PROT_EXEC | PROT_WRITE | PROT_READ,
434 };
435
436 #define OMAP_ANON 0x0002
437 #define OMAP_COPY 0x0020
438 #define OMAP_SHARED 0x0010
439 #define OMAP_FIXED 0x0100
440
441 nargs.addr = uap->addr;
442 nargs.len = uap->len;
443 nargs.prot = cvtbsdprot[uap->prot & 0x7];
444 nargs.flags = 0;
445 if (uap->flags & OMAP_ANON)
446 nargs.flags |= MAP_ANON;
447 if (uap->flags & OMAP_COPY)
448 nargs.flags |= MAP_COPY;
449 if (uap->flags & OMAP_SHARED)
450 nargs.flags |= MAP_SHARED;
451 else
452 nargs.flags |= MAP_PRIVATE;
453 if (uap->flags & OMAP_FIXED)
454 nargs.flags |= MAP_FIXED;
455 nargs.fd = uap->fd;
456 nargs.pos = uap->pos;
457 return (sys_mmap(td, &nargs));
458 }
459 #endif /* COMPAT_43 */
460
461
462 #ifndef _SYS_SYSPROTO_H_
463 struct msync_args {
464 void *addr;
465 size_t len;
466 int flags;
467 };
468 #endif
469 /*
470 * MPSAFE
471 */
472 int
473 sys_msync(td, uap)
474 struct thread *td;
475 struct msync_args *uap;
476 {
477 vm_offset_t addr;
478 vm_size_t size, pageoff;
479 int flags;
480 vm_map_t map;
481 int rv;
482
483 addr = (vm_offset_t) uap->addr;
484 size = uap->len;
485 flags = uap->flags;
486
487 pageoff = (addr & PAGE_MASK);
488 addr -= pageoff;
489 size += pageoff;
490 size = (vm_size_t) round_page(size);
491 if (addr + size < addr)
492 return (EINVAL);
493
494 if ((flags & (MS_ASYNC|MS_INVALIDATE)) == (MS_ASYNC|MS_INVALIDATE))
495 return (EINVAL);
496
497 map = &td->td_proc->p_vmspace->vm_map;
498
499 /*
500 * Clean the pages and interpret the return value.
501 */
502 rv = vm_map_sync(map, addr, addr + size, (flags & MS_ASYNC) == 0,
503 (flags & MS_INVALIDATE) != 0);
504 switch (rv) {
505 case KERN_SUCCESS:
506 return (0);
507 case KERN_INVALID_ADDRESS:
508 return (EINVAL); /* Sun returns ENOMEM? */
509 case KERN_INVALID_ARGUMENT:
510 return (EBUSY);
511 default:
512 return (EINVAL);
513 }
514 }
515
516 #ifndef _SYS_SYSPROTO_H_
517 struct munmap_args {
518 void *addr;
519 size_t len;
520 };
521 #endif
522 /*
523 * MPSAFE
524 */
525 int
526 sys_munmap(td, uap)
527 struct thread *td;
528 struct munmap_args *uap;
529 {
530 #ifdef HWPMC_HOOKS
531 struct pmckern_map_out pkm;
532 vm_map_entry_t entry;
533 #endif
534 vm_offset_t addr;
535 vm_size_t size, pageoff;
536 vm_map_t map;
537
538 addr = (vm_offset_t) uap->addr;
539 size = uap->len;
540 if (size == 0)
541 return (EINVAL);
542
543 pageoff = (addr & PAGE_MASK);
544 addr -= pageoff;
545 size += pageoff;
546 size = (vm_size_t) round_page(size);
547 if (addr + size < addr)
548 return (EINVAL);
549
550 /*
551 * Check for illegal addresses. Watch out for address wrap...
552 */
553 map = &td->td_proc->p_vmspace->vm_map;
554 if (addr < vm_map_min(map) || addr + size > vm_map_max(map))
555 return (EINVAL);
556 vm_map_lock(map);
557 #ifdef HWPMC_HOOKS
558 /*
559 * Inform hwpmc if the address range being unmapped contains
560 * an executable region.
561 */
562 pkm.pm_address = (uintptr_t) NULL;
563 if (vm_map_lookup_entry(map, addr, &entry)) {
564 for (;
565 entry != &map->header && entry->start < addr + size;
566 entry = entry->next) {
567 if (vm_map_check_protection(map, entry->start,
568 entry->end, VM_PROT_EXECUTE) == TRUE) {
569 pkm.pm_address = (uintptr_t) addr;
570 pkm.pm_size = (size_t) size;
571 break;
572 }
573 }
574 }
575 #endif
576 vm_map_delete(map, addr, addr + size);
577
578 #ifdef HWPMC_HOOKS
579 /* downgrade the lock to prevent a LOR with the pmc-sx lock */
580 vm_map_lock_downgrade(map);
581 if (pkm.pm_address != (uintptr_t) NULL)
582 PMC_CALL_HOOK(td, PMC_FN_MUNMAP, (void *) &pkm);
583 vm_map_unlock_read(map);
584 #else
585 vm_map_unlock(map);
586 #endif
587 /* vm_map_delete returns nothing but KERN_SUCCESS anyway */
588 return (0);
589 }
590
591 #ifndef _SYS_SYSPROTO_H_
592 struct mprotect_args {
593 const void *addr;
594 size_t len;
595 int prot;
596 };
597 #endif
598 /*
599 * MPSAFE
600 */
601 int
602 sys_mprotect(td, uap)
603 struct thread *td;
604 struct mprotect_args *uap;
605 {
606 vm_offset_t addr;
607 vm_size_t size, pageoff;
608 vm_prot_t prot;
609
610 addr = (vm_offset_t) uap->addr;
611 size = uap->len;
612 prot = uap->prot & VM_PROT_ALL;
613
614 pageoff = (addr & PAGE_MASK);
615 addr -= pageoff;
616 size += pageoff;
617 size = (vm_size_t) round_page(size);
618 if (addr + size < addr)
619 return (EINVAL);
620
621 switch (vm_map_protect(&td->td_proc->p_vmspace->vm_map, addr,
622 addr + size, prot, FALSE)) {
623 case KERN_SUCCESS:
624 return (0);
625 case KERN_PROTECTION_FAILURE:
626 return (EACCES);
627 case KERN_RESOURCE_SHORTAGE:
628 return (ENOMEM);
629 }
630 return (EINVAL);
631 }
632
633 #ifndef _SYS_SYSPROTO_H_
634 struct minherit_args {
635 void *addr;
636 size_t len;
637 int inherit;
638 };
639 #endif
640 /*
641 * MPSAFE
642 */
643 int
644 sys_minherit(td, uap)
645 struct thread *td;
646 struct minherit_args *uap;
647 {
648 vm_offset_t addr;
649 vm_size_t size, pageoff;
650 vm_inherit_t inherit;
651
652 addr = (vm_offset_t)uap->addr;
653 size = uap->len;
654 inherit = uap->inherit;
655
656 pageoff = (addr & PAGE_MASK);
657 addr -= pageoff;
658 size += pageoff;
659 size = (vm_size_t) round_page(size);
660 if (addr + size < addr)
661 return (EINVAL);
662
663 switch (vm_map_inherit(&td->td_proc->p_vmspace->vm_map, addr,
664 addr + size, inherit)) {
665 case KERN_SUCCESS:
666 return (0);
667 case KERN_PROTECTION_FAILURE:
668 return (EACCES);
669 }
670 return (EINVAL);
671 }
672
673 #ifndef _SYS_SYSPROTO_H_
674 struct madvise_args {
675 void *addr;
676 size_t len;
677 int behav;
678 };
679 #endif
680
681 /*
682 * MPSAFE
683 */
684 /* ARGSUSED */
685 int
686 sys_madvise(td, uap)
687 struct thread *td;
688 struct madvise_args *uap;
689 {
690 vm_offset_t start, end;
691 vm_map_t map;
692 struct proc *p;
693 int error;
694
695 /*
696 * Check for our special case, advising the swap pager we are
697 * "immortal."
698 */
699 if (uap->behav == MADV_PROTECT) {
700 error = priv_check(td, PRIV_VM_MADV_PROTECT);
701 if (error == 0) {
702 p = td->td_proc;
703 PROC_LOCK(p);
704 p->p_flag |= P_PROTECTED;
705 PROC_UNLOCK(p);
706 }
707 return (error);
708 }
709 /*
710 * Check for illegal behavior
711 */
712 if (uap->behav < 0 || uap->behav > MADV_CORE)
713 return (EINVAL);
714 /*
715 * Check for illegal addresses. Watch out for address wrap... Note
716 * that VM_*_ADDRESS are not constants due to casts (argh).
717 */
718 map = &td->td_proc->p_vmspace->vm_map;
719 if ((vm_offset_t)uap->addr < vm_map_min(map) ||
720 (vm_offset_t)uap->addr + uap->len > vm_map_max(map))
721 return (EINVAL);
722 if (((vm_offset_t) uap->addr + uap->len) < (vm_offset_t) uap->addr)
723 return (EINVAL);
724
725 /*
726 * Since this routine is only advisory, we default to conservative
727 * behavior.
728 */
729 start = trunc_page((vm_offset_t) uap->addr);
730 end = round_page((vm_offset_t) uap->addr + uap->len);
731
732 if (vm_map_madvise(map, start, end, uap->behav))
733 return (EINVAL);
734 return (0);
735 }
736
737 #ifndef _SYS_SYSPROTO_H_
738 struct mincore_args {
739 const void *addr;
740 size_t len;
741 char *vec;
742 };
743 #endif
744
745 /*
746 * MPSAFE
747 */
748 /* ARGSUSED */
749 int
750 sys_mincore(td, uap)
751 struct thread *td;
752 struct mincore_args *uap;
753 {
754 vm_offset_t addr, first_addr;
755 vm_offset_t end, cend;
756 pmap_t pmap;
757 vm_map_t map;
758 char *vec;
759 int error = 0;
760 int vecindex, lastvecindex;
761 vm_map_entry_t current;
762 vm_map_entry_t entry;
763 vm_object_t object;
764 vm_paddr_t locked_pa;
765 vm_page_t m;
766 vm_pindex_t pindex;
767 int mincoreinfo;
768 unsigned int timestamp;
769 boolean_t locked;
770
771 /*
772 * Make sure that the addresses presented are valid for user
773 * mode.
774 */
775 first_addr = addr = trunc_page((vm_offset_t) uap->addr);
776 end = addr + (vm_size_t)round_page(uap->len);
777 map = &td->td_proc->p_vmspace->vm_map;
778 if (end > vm_map_max(map) || end < addr)
779 return (ENOMEM);
780
781 /*
782 * Address of byte vector
783 */
784 vec = uap->vec;
785
786 pmap = vmspace_pmap(td->td_proc->p_vmspace);
787
788 vm_map_lock_read(map);
789 RestartScan:
790 timestamp = map->timestamp;
791
792 if (!vm_map_lookup_entry(map, addr, &entry)) {
793 vm_map_unlock_read(map);
794 return (ENOMEM);
795 }
796
797 /*
798 * Do this on a map entry basis so that if the pages are not
799 * in the current processes address space, we can easily look
800 * up the pages elsewhere.
801 */
802 lastvecindex = -1;
803 for (current = entry;
804 (current != &map->header) && (current->start < end);
805 current = current->next) {
806
807 /*
808 * check for contiguity
809 */
810 if (current->end < end &&
811 (entry->next == &map->header ||
812 current->next->start > current->end)) {
813 vm_map_unlock_read(map);
814 return (ENOMEM);
815 }
816
817 /*
818 * ignore submaps (for now) or null objects
819 */
820 if ((current->eflags & MAP_ENTRY_IS_SUB_MAP) ||
821 current->object.vm_object == NULL)
822 continue;
823
824 /*
825 * limit this scan to the current map entry and the
826 * limits for the mincore call
827 */
828 if (addr < current->start)
829 addr = current->start;
830 cend = current->end;
831 if (cend > end)
832 cend = end;
833
834 /*
835 * scan this entry one page at a time
836 */
837 while (addr < cend) {
838 /*
839 * Check pmap first, it is likely faster, also
840 * it can provide info as to whether we are the
841 * one referencing or modifying the page.
842 */
843 object = NULL;
844 locked_pa = 0;
845 retry:
846 m = NULL;
847 mincoreinfo = pmap_mincore(pmap, addr, &locked_pa);
848 if (locked_pa != 0) {
849 /*
850 * The page is mapped by this process but not
851 * both accessed and modified. It is also
852 * managed. Acquire the object lock so that
853 * other mappings might be examined.
854 */
855 m = PHYS_TO_VM_PAGE(locked_pa);
856 if (m->object != object) {
857 if (object != NULL)
858 VM_OBJECT_UNLOCK(object);
859 object = m->object;
860 locked = VM_OBJECT_TRYLOCK(object);
861 vm_page_unlock(m);
862 if (!locked) {
863 VM_OBJECT_LOCK(object);
864 vm_page_lock(m);
865 goto retry;
866 }
867 } else
868 vm_page_unlock(m);
869 KASSERT(m->valid == VM_PAGE_BITS_ALL,
870 ("mincore: page %p is mapped but invalid",
871 m));
872 } else if (mincoreinfo == 0) {
873 /*
874 * The page is not mapped by this process. If
875 * the object implements managed pages, then
876 * determine if the page is resident so that
877 * the mappings might be examined.
878 */
879 if (current->object.vm_object != object) {
880 if (object != NULL)
881 VM_OBJECT_UNLOCK(object);
882 object = current->object.vm_object;
883 VM_OBJECT_LOCK(object);
884 }
885 if (object->type == OBJT_DEFAULT ||
886 object->type == OBJT_SWAP ||
887 object->type == OBJT_VNODE) {
888 pindex = OFF_TO_IDX(current->offset +
889 (addr - current->start));
890 m = vm_page_lookup(object, pindex);
891 if (m != NULL && m->valid == 0)
892 m = NULL;
893 if (m != NULL)
894 mincoreinfo = MINCORE_INCORE;
895 }
896 }
897 if (m != NULL) {
898 /* Examine other mappings to the page. */
899 if (m->dirty == 0 && pmap_is_modified(m))
900 vm_page_dirty(m);
901 if (m->dirty != 0)
902 mincoreinfo |= MINCORE_MODIFIED_OTHER;
903 /*
904 * The first test for PGA_REFERENCED is an
905 * optimization. The second test is
906 * required because a concurrent pmap
907 * operation could clear the last reference
908 * and set PGA_REFERENCED before the call to
909 * pmap_is_referenced().
910 */
911 if ((m->aflags & PGA_REFERENCED) != 0 ||
912 pmap_is_referenced(m) ||
913 (m->aflags & PGA_REFERENCED) != 0)
914 mincoreinfo |= MINCORE_REFERENCED_OTHER;
915 }
916 if (object != NULL)
917 VM_OBJECT_UNLOCK(object);
918
919 /*
920 * subyte may page fault. In case it needs to modify
921 * the map, we release the lock.
922 */
923 vm_map_unlock_read(map);
924
925 /*
926 * calculate index into user supplied byte vector
927 */
928 vecindex = OFF_TO_IDX(addr - first_addr);
929
930 /*
931 * If we have skipped map entries, we need to make sure that
932 * the byte vector is zeroed for those skipped entries.
933 */
934 while ((lastvecindex + 1) < vecindex) {
935 error = subyte(vec + lastvecindex, 0);
936 if (error) {
937 error = EFAULT;
938 goto done2;
939 }
940 ++lastvecindex;
941 }
942
943 /*
944 * Pass the page information to the user
945 */
946 error = subyte(vec + vecindex, mincoreinfo);
947 if (error) {
948 error = EFAULT;
949 goto done2;
950 }
951
952 /*
953 * If the map has changed, due to the subyte, the previous
954 * output may be invalid.
955 */
956 vm_map_lock_read(map);
957 if (timestamp != map->timestamp)
958 goto RestartScan;
959
960 lastvecindex = vecindex;
961 addr += PAGE_SIZE;
962 }
963 }
964
965 /*
966 * subyte may page fault. In case it needs to modify
967 * the map, we release the lock.
968 */
969 vm_map_unlock_read(map);
970
971 /*
972 * Zero the last entries in the byte vector.
973 */
974 vecindex = OFF_TO_IDX(end - first_addr);
975 while ((lastvecindex + 1) < vecindex) {
976 error = subyte(vec + lastvecindex, 0);
977 if (error) {
978 error = EFAULT;
979 goto done2;
980 }
981 ++lastvecindex;
982 }
983
984 /*
985 * If the map has changed, due to the subyte, the previous
986 * output may be invalid.
987 */
988 vm_map_lock_read(map);
989 if (timestamp != map->timestamp)
990 goto RestartScan;
991 vm_map_unlock_read(map);
992 done2:
993 return (error);
994 }
995
996 #ifndef _SYS_SYSPROTO_H_
997 struct mlock_args {
998 const void *addr;
999 size_t len;
1000 };
1001 #endif
1002 /*
1003 * MPSAFE
1004 */
1005 int
1006 sys_mlock(td, uap)
1007 struct thread *td;
1008 struct mlock_args *uap;
1009 {
1010 struct proc *proc;
1011 vm_offset_t addr, end, last, start;
1012 vm_size_t npages, size;
1013 unsigned long nsize;
1014 int error;
1015
1016 error = priv_check(td, PRIV_VM_MLOCK);
1017 if (error)
1018 return (error);
1019 addr = (vm_offset_t)uap->addr;
1020 size = uap->len;
1021 last = addr + size;
1022 start = trunc_page(addr);
1023 end = round_page(last);
1024 if (last < addr || end < addr)
1025 return (EINVAL);
1026 npages = atop(end - start);
1027 if (npages > vm_page_max_wired)
1028 return (ENOMEM);
1029 proc = td->td_proc;
1030 PROC_LOCK(proc);
1031 nsize = ptoa(npages +
1032 pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map)));
1033 if (nsize > lim_cur(proc, RLIMIT_MEMLOCK)) {
1034 PROC_UNLOCK(proc);
1035 return (ENOMEM);
1036 }
1037 PROC_UNLOCK(proc);
1038 if (npages + cnt.v_wire_count > vm_page_max_wired)
1039 return (EAGAIN);
1040 #ifdef RACCT
1041 PROC_LOCK(proc);
1042 error = racct_set(proc, RACCT_MEMLOCK, nsize);
1043 PROC_UNLOCK(proc);
1044 if (error != 0)
1045 return (ENOMEM);
1046 #endif
1047 error = vm_map_wire(&proc->p_vmspace->vm_map, start, end,
1048 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1049 #ifdef RACCT
1050 if (error != KERN_SUCCESS) {
1051 PROC_LOCK(proc);
1052 racct_set(proc, RACCT_MEMLOCK,
1053 ptoa(pmap_wired_count(vm_map_pmap(&proc->p_vmspace->vm_map))));
1054 PROC_UNLOCK(proc);
1055 }
1056 #endif
1057 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1058 }
1059
1060 #ifndef _SYS_SYSPROTO_H_
1061 struct mlockall_args {
1062 int how;
1063 };
1064 #endif
1065
1066 /*
1067 * MPSAFE
1068 */
1069 int
1070 sys_mlockall(td, uap)
1071 struct thread *td;
1072 struct mlockall_args *uap;
1073 {
1074 vm_map_t map;
1075 int error;
1076
1077 map = &td->td_proc->p_vmspace->vm_map;
1078 error = 0;
1079
1080 if ((uap->how == 0) || ((uap->how & ~(MCL_CURRENT|MCL_FUTURE)) != 0))
1081 return (EINVAL);
1082
1083 #if 0
1084 /*
1085 * If wiring all pages in the process would cause it to exceed
1086 * a hard resource limit, return ENOMEM.
1087 */
1088 PROC_LOCK(td->td_proc);
1089 if (map->size > lim_cur(td->td_proc, RLIMIT_MEMLOCK)) {
1090 PROC_UNLOCK(td->td_proc);
1091 return (ENOMEM);
1092 }
1093 PROC_UNLOCK(td->td_proc);
1094 #else
1095 error = priv_check(td, PRIV_VM_MLOCK);
1096 if (error)
1097 return (error);
1098 #endif
1099 #ifdef RACCT
1100 PROC_LOCK(td->td_proc);
1101 error = racct_set(td->td_proc, RACCT_MEMLOCK, map->size);
1102 PROC_UNLOCK(td->td_proc);
1103 if (error != 0)
1104 return (ENOMEM);
1105 #endif
1106
1107 if (uap->how & MCL_FUTURE) {
1108 vm_map_lock(map);
1109 vm_map_modflags(map, MAP_WIREFUTURE, 0);
1110 vm_map_unlock(map);
1111 error = 0;
1112 }
1113
1114 if (uap->how & MCL_CURRENT) {
1115 /*
1116 * P1003.1-2001 mandates that all currently mapped pages
1117 * will be memory resident and locked (wired) upon return
1118 * from mlockall(). vm_map_wire() will wire pages, by
1119 * calling vm_fault_wire() for each page in the region.
1120 */
1121 error = vm_map_wire(map, vm_map_min(map), vm_map_max(map),
1122 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1123 error = (error == KERN_SUCCESS ? 0 : EAGAIN);
1124 }
1125 #ifdef RACCT
1126 if (error != KERN_SUCCESS) {
1127 PROC_LOCK(td->td_proc);
1128 racct_set(td->td_proc, RACCT_MEMLOCK,
1129 ptoa(pmap_wired_count(vm_map_pmap(&td->td_proc->p_vmspace->vm_map))));
1130 PROC_UNLOCK(td->td_proc);
1131 }
1132 #endif
1133
1134 return (error);
1135 }
1136
1137 #ifndef _SYS_SYSPROTO_H_
1138 struct munlockall_args {
1139 register_t dummy;
1140 };
1141 #endif
1142
1143 /*
1144 * MPSAFE
1145 */
1146 int
1147 sys_munlockall(td, uap)
1148 struct thread *td;
1149 struct munlockall_args *uap;
1150 {
1151 vm_map_t map;
1152 int error;
1153
1154 map = &td->td_proc->p_vmspace->vm_map;
1155 error = priv_check(td, PRIV_VM_MUNLOCK);
1156 if (error)
1157 return (error);
1158
1159 /* Clear the MAP_WIREFUTURE flag from this vm_map. */
1160 vm_map_lock(map);
1161 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1162 vm_map_unlock(map);
1163
1164 /* Forcibly unwire all pages. */
1165 error = vm_map_unwire(map, vm_map_min(map), vm_map_max(map),
1166 VM_MAP_WIRE_USER|VM_MAP_WIRE_HOLESOK);
1167 #ifdef RACCT
1168 if (error == KERN_SUCCESS) {
1169 PROC_LOCK(td->td_proc);
1170 racct_set(td->td_proc, RACCT_MEMLOCK, 0);
1171 PROC_UNLOCK(td->td_proc);
1172 }
1173 #endif
1174
1175 return (error);
1176 }
1177
1178 #ifndef _SYS_SYSPROTO_H_
1179 struct munlock_args {
1180 const void *addr;
1181 size_t len;
1182 };
1183 #endif
1184 /*
1185 * MPSAFE
1186 */
1187 int
1188 sys_munlock(td, uap)
1189 struct thread *td;
1190 struct munlock_args *uap;
1191 {
1192 vm_offset_t addr, end, last, start;
1193 vm_size_t size;
1194 int error;
1195
1196 error = priv_check(td, PRIV_VM_MUNLOCK);
1197 if (error)
1198 return (error);
1199 addr = (vm_offset_t)uap->addr;
1200 size = uap->len;
1201 last = addr + size;
1202 start = trunc_page(addr);
1203 end = round_page(last);
1204 if (last < addr || end < addr)
1205 return (EINVAL);
1206 error = vm_map_unwire(&td->td_proc->p_vmspace->vm_map, start, end,
1207 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
1208 #ifdef RACCT
1209 if (error == KERN_SUCCESS) {
1210 PROC_LOCK(td->td_proc);
1211 racct_sub(td->td_proc, RACCT_MEMLOCK, ptoa(end - start));
1212 PROC_UNLOCK(td->td_proc);
1213 }
1214 #endif
1215 return (error == KERN_SUCCESS ? 0 : ENOMEM);
1216 }
1217
1218 /*
1219 * vm_mmap_vnode()
1220 *
1221 * MPSAFE
1222 *
1223 * Helper function for vm_mmap. Perform sanity check specific for mmap
1224 * operations on vnodes.
1225 */
1226 int
1227 vm_mmap_vnode(struct thread *td, vm_size_t objsize,
1228 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1229 struct vnode *vp, vm_ooffset_t *foffp, vm_object_t *objp)
1230 {
1231 struct vattr va;
1232 vm_object_t obj;
1233 vm_offset_t foff;
1234 struct mount *mp;
1235 struct ucred *cred;
1236 int error, flags;
1237 int vfslocked;
1238
1239 mp = vp->v_mount;
1240 cred = td->td_ucred;
1241 vfslocked = VFS_LOCK_GIANT(mp);
1242 if ((error = vget(vp, LK_SHARED, td)) != 0) {
1243 VFS_UNLOCK_GIANT(vfslocked);
1244 return (error);
1245 }
1246 foff = *foffp;
1247 flags = *flagsp;
1248 obj = vp->v_object;
1249 if (vp->v_type == VREG) {
1250 /*
1251 * Get the proper underlying object
1252 */
1253 if (obj == NULL) {
1254 error = EINVAL;
1255 goto done;
1256 }
1257 if (obj->handle != vp) {
1258 vput(vp);
1259 vp = (struct vnode*)obj->handle;
1260 vget(vp, LK_SHARED, td);
1261 }
1262 } else if (vp->v_type == VCHR) {
1263 error = vm_mmap_cdev(td, objsize, prot, maxprotp, flagsp,
1264 vp->v_rdev, foffp, objp);
1265 if (error == 0)
1266 goto mark_atime;
1267 goto done;
1268 } else {
1269 error = EINVAL;
1270 goto done;
1271 }
1272 if ((error = VOP_GETATTR(vp, &va, cred)))
1273 goto done;
1274 #ifdef MAC
1275 error = mac_vnode_check_mmap(cred, vp, prot, flags);
1276 if (error != 0)
1277 goto done;
1278 #endif
1279 if ((flags & MAP_SHARED) != 0) {
1280 if ((va.va_flags & (SF_SNAPSHOT|IMMUTABLE|APPEND)) != 0) {
1281 if (prot & PROT_WRITE) {
1282 error = EPERM;
1283 goto done;
1284 }
1285 *maxprotp &= ~VM_PROT_WRITE;
1286 }
1287 }
1288 /*
1289 * If it is a regular file without any references
1290 * we do not need to sync it.
1291 * Adjust object size to be the size of actual file.
1292 */
1293 objsize = round_page(va.va_size);
1294 if (va.va_nlink == 0)
1295 flags |= MAP_NOSYNC;
1296 obj = vm_pager_allocate(OBJT_VNODE, vp, objsize, prot, foff, td->td_ucred);
1297 if (obj == NULL) {
1298 error = ENOMEM;
1299 goto done;
1300 }
1301 *objp = obj;
1302 *flagsp = flags;
1303
1304 mark_atime:
1305 vfs_mark_atime(vp, cred);
1306
1307 done:
1308 vput(vp);
1309 VFS_UNLOCK_GIANT(vfslocked);
1310 return (error);
1311 }
1312
1313 /*
1314 * vm_mmap_cdev()
1315 *
1316 * MPSAFE
1317 *
1318 * Helper function for vm_mmap. Perform sanity check specific for mmap
1319 * operations on cdevs.
1320 */
1321 int
1322 vm_mmap_cdev(struct thread *td, vm_size_t objsize,
1323 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1324 struct cdev *cdev, vm_ooffset_t *foff, vm_object_t *objp)
1325 {
1326 vm_object_t obj;
1327 struct cdevsw *dsw;
1328 int error, flags, ref;
1329
1330 flags = *flagsp;
1331
1332 dsw = dev_refthread(cdev, &ref);
1333 if (dsw == NULL)
1334 return (ENXIO);
1335 if (dsw->d_flags & D_MMAP_ANON) {
1336 dev_relthread(cdev, ref);
1337 *maxprotp = VM_PROT_ALL;
1338 *flagsp |= MAP_ANON;
1339 return (0);
1340 }
1341 /*
1342 * cdevs do not provide private mappings of any kind.
1343 */
1344 if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1345 (prot & PROT_WRITE) != 0) {
1346 dev_relthread(cdev, ref);
1347 return (EACCES);
1348 }
1349 if (flags & (MAP_PRIVATE|MAP_COPY)) {
1350 dev_relthread(cdev, ref);
1351 return (EINVAL);
1352 }
1353 /*
1354 * Force device mappings to be shared.
1355 */
1356 flags |= MAP_SHARED;
1357 #ifdef MAC_XXX
1358 error = mac_cdev_check_mmap(td->td_ucred, cdev, prot);
1359 if (error != 0) {
1360 dev_relthread(cdev, ref);
1361 return (error);
1362 }
1363 #endif
1364 /*
1365 * First, try d_mmap_single(). If that is not implemented
1366 * (returns ENODEV), fall back to using the device pager.
1367 * Note that d_mmap_single() must return a reference to the
1368 * object (it needs to bump the reference count of the object
1369 * it returns somehow).
1370 *
1371 * XXX assumes VM_PROT_* == PROT_*
1372 */
1373 error = dsw->d_mmap_single(cdev, foff, objsize, objp, (int)prot);
1374 dev_relthread(cdev, ref);
1375 if (error != ENODEV)
1376 return (error);
1377 obj = vm_pager_allocate(OBJT_DEVICE, cdev, objsize, prot, *foff,
1378 td->td_ucred);
1379 if (obj == NULL)
1380 return (EINVAL);
1381 *objp = obj;
1382 *flagsp = flags;
1383 return (0);
1384 }
1385
1386 /*
1387 * vm_mmap_shm()
1388 *
1389 * MPSAFE
1390 *
1391 * Helper function for vm_mmap. Perform sanity check specific for mmap
1392 * operations on shm file descriptors.
1393 */
1394 int
1395 vm_mmap_shm(struct thread *td, vm_size_t objsize,
1396 vm_prot_t prot, vm_prot_t *maxprotp, int *flagsp,
1397 struct shmfd *shmfd, vm_ooffset_t foff, vm_object_t *objp)
1398 {
1399 int error;
1400
1401 if ((*flagsp & MAP_SHARED) != 0 &&
1402 (*maxprotp & VM_PROT_WRITE) == 0 &&
1403 (prot & PROT_WRITE) != 0)
1404 return (EACCES);
1405 #ifdef MAC
1406 error = mac_posixshm_check_mmap(td->td_ucred, shmfd, prot, *flagsp);
1407 if (error != 0)
1408 return (error);
1409 #endif
1410 error = shm_mmap(shmfd, objsize, foff, objp);
1411 if (error)
1412 return (error);
1413 return (0);
1414 }
1415
1416 /*
1417 * vm_mmap()
1418 *
1419 * MPSAFE
1420 *
1421 * Internal version of mmap. Currently used by mmap, exec, and sys5
1422 * shared memory. Handle is either a vnode pointer or NULL for MAP_ANON.
1423 */
1424 int
1425 vm_mmap(vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot,
1426 vm_prot_t maxprot, int flags,
1427 objtype_t handle_type, void *handle,
1428 vm_ooffset_t foff)
1429 {
1430 boolean_t fitit;
1431 vm_object_t object = NULL;
1432 int rv = KERN_SUCCESS;
1433 int docow, error;
1434 struct thread *td = curthread;
1435
1436 if (size == 0)
1437 return (0);
1438
1439 size = round_page(size);
1440
1441 PROC_LOCK(td->td_proc);
1442 if (td->td_proc->p_vmspace->vm_map.size + size >
1443 lim_cur(td->td_proc, RLIMIT_VMEM)) {
1444 PROC_UNLOCK(td->td_proc);
1445 return (ENOMEM);
1446 }
1447 if (racct_set(td->td_proc, RACCT_VMEM,
1448 td->td_proc->p_vmspace->vm_map.size + size)) {
1449 PROC_UNLOCK(td->td_proc);
1450 return (ENOMEM);
1451 }
1452 PROC_UNLOCK(td->td_proc);
1453
1454 /*
1455 * We currently can only deal with page aligned file offsets.
1456 * The check is here rather than in the syscall because the
1457 * kernel calls this function internally for other mmaping
1458 * operations (such as in exec) and non-aligned offsets will
1459 * cause pmap inconsistencies...so we want to be sure to
1460 * disallow this in all cases.
1461 */
1462 if (foff & PAGE_MASK)
1463 return (EINVAL);
1464
1465 if ((flags & MAP_FIXED) == 0) {
1466 fitit = TRUE;
1467 *addr = round_page(*addr);
1468 } else {
1469 if (*addr != trunc_page(*addr))
1470 return (EINVAL);
1471 fitit = FALSE;
1472 }
1473 /*
1474 * Lookup/allocate object.
1475 */
1476 switch (handle_type) {
1477 case OBJT_DEVICE:
1478 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags,
1479 handle, &foff, &object);
1480 break;
1481 case OBJT_VNODE:
1482 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags,
1483 handle, &foff, &object);
1484 break;
1485 case OBJT_SWAP:
1486 error = vm_mmap_shm(td, size, prot, &maxprot, &flags,
1487 handle, foff, &object);
1488 break;
1489 case OBJT_DEFAULT:
1490 if (handle == NULL) {
1491 error = 0;
1492 break;
1493 }
1494 /* FALLTHROUGH */
1495 default:
1496 error = EINVAL;
1497 break;
1498 }
1499 if (error)
1500 return (error);
1501 if (flags & MAP_ANON) {
1502 object = NULL;
1503 docow = 0;
1504 /*
1505 * Unnamed anonymous regions always start at 0.
1506 */
1507 if (handle == 0)
1508 foff = 0;
1509 } else if (flags & MAP_PREFAULT_READ)
1510 docow = MAP_PREFAULT;
1511 else
1512 docow = MAP_PREFAULT_PARTIAL;
1513
1514 if ((flags & (MAP_ANON|MAP_SHARED)) == 0)
1515 docow |= MAP_COPY_ON_WRITE;
1516 if (flags & MAP_NOSYNC)
1517 docow |= MAP_DISABLE_SYNCER;
1518 if (flags & MAP_NOCORE)
1519 docow |= MAP_DISABLE_COREDUMP;
1520
1521 if (flags & MAP_STACK)
1522 rv = vm_map_stack(map, *addr, size, prot, maxprot,
1523 docow | MAP_STACK_GROWS_DOWN);
1524 else if (fitit)
1525 rv = vm_map_find(map, object, foff, addr, size,
1526 object != NULL && object->type == OBJT_DEVICE ?
1527 VMFS_ALIGNED_SPACE : VMFS_ANY_SPACE, prot, maxprot, docow);
1528 else
1529 rv = vm_map_fixed(map, object, foff, *addr, size,
1530 prot, maxprot, docow);
1531
1532 if (rv != KERN_SUCCESS) {
1533 /*
1534 * Lose the object reference. Will destroy the
1535 * object if it's an unnamed anonymous mapping
1536 * or named anonymous without other references.
1537 */
1538 vm_object_deallocate(object);
1539 } else if (flags & MAP_SHARED) {
1540 /*
1541 * Shared memory is also shared with children.
1542 */
1543 rv = vm_map_inherit(map, *addr, *addr + size, VM_INHERIT_SHARE);
1544 if (rv != KERN_SUCCESS)
1545 (void) vm_map_remove(map, *addr, *addr + size);
1546 }
1547
1548 /*
1549 * If the process has requested that all future mappings
1550 * be wired, then heed this.
1551 */
1552 if ((rv == KERN_SUCCESS) && (map->flags & MAP_WIREFUTURE))
1553 vm_map_wire(map, *addr, *addr + size,
1554 VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
1555
1556 return (vm_mmap_to_errno(rv));
1557 }
1558
1559 int
1560 vm_mmap_to_errno(int rv)
1561 {
1562
1563 switch (rv) {
1564 case KERN_SUCCESS:
1565 return (0);
1566 case KERN_INVALID_ADDRESS:
1567 case KERN_NO_SPACE:
1568 return (ENOMEM);
1569 case KERN_PROTECTION_FAILURE:
1570 return (EACCES);
1571 default:
1572 return (EINVAL);
1573 }
1574 }
Cache object: 9e13c1e574fe94177833e448eb5636ef
|